chore(server): quiet benign Gemma4-Assistant memory-probe warning on MTP load

[marksverdhei]

Summary

• The MTP draft memory-probe in server-context.cpp (~line 856) throws on Gemma4-Assistant because cparams.ctx_other is required and the target context doesn't exist yet at probe time.
• Upstream's own init explicitly notes "this is normal during memory fitting" in the exception message and carries a TODO to switch to a typed llama_exception so the warning can be skipped.
• Until that upstream change lands and flows in via a master sync, scan the exception message for the self-identifying "normal during memory fitting" marker and downgrade SRV_WRN -> SRV_DBG for that specific case. Real failures (model load failed, etc.) still surface as SRV_WRN.

Why

• Eliminates the misleading [spec] failed to measure draft model memory: failed to create llama_context from model line that appears on every gemma-4-12b-qat-mtp pod start despite the model loading + running fine at ~110 tok/s (verified Phase 6 on titan, image unified-llm:mtp-pr23398-5e6dff22).
• The warning was non-blocking and we shipped γ around it, but it pollutes the logs every load and would lead to a future operator chasing a non-bug.

Test plan

• cmake build of llama-server still clean (CUDA on, all targets)
• No behavior change for real failures (only downgrade when the exception self-identifies as benign-during-fit)
• Snoop verifies the warning is gone on the next titan pod restart (no roll/bake needed — just a logging change; we'll fold it into the next bake naturally)

Related

• Source of the exception: src/llama-context.cpp near the GEMMA4_ASSISTANT ctx_other check (carries a TODO from am17an's upstream PR llama : add Gemma4 MTP ggml-org/llama.cpp#23398 to switch to typed exception)
• γ master sync PR that brought the warning in: feat(sync): upstream master sync (42 commits) + Gemma4 MTP via PR #23398 vendor #93